🔧 Apply audio flow fixes to disabled language outputs

[ajlennon]

Overview

When re-enabling currently disabled language outputs, they will need the same audio flow improvements that were applied to en_GB, pl_PL, and bella.

Disabled Languages

The following languages are currently disabled for cost optimization:

• fr_FR (French)
• de_DE (German)
• es_ES (Spanish)
• it_IT (Italian)
• nl_NL (Dutch)
• en_GB_LON (English - London)
• en_GB_LIV (English - Liverpool)

Required Updates

1. AI Prompt Updates

Each language's synthesis prompt in config/ai_prompts.json needs:

• CRITICAL FLOW RULES section added (similar to en_GB/pl_PL/bella)
• Instructions to use semicolons/commas instead of periods for transitions
• Lowercase section starts when continuing from previous sentence
• Minimize comma usage for smoother flow
• No quote marks

2. Post-Processing Fixes

Add language-specific transition pattern fixes in scripts/github_ai_news_digest.py:

• Identify common section transition phrases for each language
• Add regex patterns to replace periods with semicolons/commas before transitions
• Example patterns:
  • French: "Dans l'actualité {theme}..."
  • German: "In den {theme}-Nachrichten..."
  • Spanish: "En las noticias de {theme}..."
  • Italian: "Nelle notizie {theme}..."
  • Dutch: "In het {theme}-nieuws..."

3. Testing

When re-enabling:

1. Generate test content for the language
2. Verify speech rate is ~150 WPM (not <120 WPM)
3. Check for unnatural pauses (>0.8s) using audio analysis
4. Verify section transitions flow smoothly without breaks

Reference

See commits:

• 01246f5 - Initial flow fixes for bella
• 6bc5e04 - Applied fixes to en_GB and pl_PL

Priority

Low - Only needed when/if these languages are re-enabled.